Optimize computations by working with `MPolyRingElem` instead of `UPoly` where possible

[fingolfin]

We already discussed this before, just to make sure it is not forgotten:

When possible we should consider unpacking all UPoly into MPolyRingElem then working with those, then only wrapping the result back into UPoly. That should avoid redundant allocations and also in some cases ensure we use optimized algorithms.

Caveat: we need to make sure the unwrapped polynomials all have the same parent. There is no nice helper for this right now, but there should be... in the meantime we can write something like this (untested, may need tweaking, but gives the right idea):

function unwrap(a)
   R = parent(a)
   s = data(a)
   if nvars(parent(s)) != nvars(R)
      s = upgrade(R, s)
   end
   return s
end

[SoongNoonien]

I've opened an issue over at AbstractAlgebra about this unwrapping: Nemocas/AbstractAlgebra.jl#1969